Google Lens Study: Results From 65,388 Visual Searches

We analyzed 65,388 Google Lens search results to better understand how visual search works.

Specifically, we looked at potential ranking factors that Google may use in its Lens algorithm, including:

  • Alt text
  • Domain Authority
  • URLs
  • Title tags
  • Responsive design

And in this post I’m going to share what we discovered.

A Summary of Our Most Interesting Findings is As Follows:

1. 32.5% of all Google Lens results have a “matching” keyword in the page’s title tag. Therefore, keyword-optimized title tags may help a page rank in Google Lens.

2. A high proportion of Google Lens results are pulled from images high up on a page. In fact, approximately 1/3rd of all Google Lens image results appear in the top 25% of a web page.

3. 11.4% of all Google Lens result images contain alt text terms that match the keyword someone just searched for.

4. Authoritative pages and websites appear to have a ranking advantage in Google Lens. Google Lens results have an average Moz Page Authority of 35.2 and a Domain Authority of 64.4.

5. Google Lens may confer a ranking advantage to sites that work well on mobile devices. 90.6% of all Google Lens results come from mobile-friendly websites.

6. Google Lens results come from relatively slow-loading pages. In fact, the average First Contentful Paint load time for a Google Lens result is 3,186ms.

7. Pinterest and Amazon are the two top-performing websites in Google Lens. 7.2% of all Google Lens results come from Pinterest, while 4.1% are from Amazon.

8. Keyword-rich URLs correlate with Google Lens rankings. 29.9% of the Google Lens results in our analysis had an associated keyword in the page’s URL.

9. Responsive images don’t appear to confer a major ranking advantage in Google Lens. Only 13.1% of Google Lens images are responsive.

10. Images with filenames that “match” a Google Lens search appear to rank more often than images with a missing or irrelevant filename. We found that 22.6% of Google Lens result images had a filename that matched the Google Lens search.

11. Websites that rank seem to have an edge in Google Lens. We discovered that 15% of all Google Lens results also rank on the first page of Google organic for the same search term.

12. The average visual search result page contains 1,631 words. Considering that Google uses text around an image to understand an image’s content and context, this finding makes sense.

Keyword-Rich Title Tags and URLs Correlate With Google Lens Rankings

With Google Lens, the image that you search with is your “keyword”.

We used the Google Vision API to turn those visual searches into text-based keywords.

Google Vision API – Web Entities Empire State Building

This allowed us to analyze visual search terms as text-based entities (using labels from Google themselves).

For example, take this Google Lens “keyword”:

Example Search With Image Of Cat

The Google Vision API turned that image into a “best guess label”: “cat”.

Google Vision API – Cat Result

We then looked at whether that text-based label correlated with potential Google Lens ranking factors.

In fact, that’s how we were able to correlate title tag keywords and Google Lens rankings.

We found 32.5% of pages that rank in Google Lens have a keyword in their title tag that matches the search image’s Google Vision label.

Pages That Rank In Google Lens Have A Keyword That Matches The Search Images Google Vision Label

For example, here is one of the visual searches from our dataset.

Example Visual Search – Tesla Cybertruck

Sure enough, Google Vision’s label (“tesla cybertruck”), appears in the title tag for that page.

Tesla Cybertruck In Title Tag

Google has previously stated that they use a page’s title tag to rank Google Images results.

Google Use Page Title Tag For Lens Rankings

It would make sense that they’d also use keywords in a page’s title tag for Google Lens rankings.

In other words, Google may prefer images on pages “about” that image vs. pages that have a matching image in the middle of an otherwise irrelevant page.

Google Prefers Images On Pages About That Image

Indeed, that’s what our data showed.

Key Takeaway: Pages with titles and URLs that “match” a Google Lens search term may have a rankings advantage over those that don’t. With visual search, the image that you search with is your “keyword”.

33% of Google Lens Result Images Appear Towards The Top of a Page

We found that 33.1% Google Lens results are from images from the top 25% of a webpage.

33% Of Google Lens Results Are From Images From The Top 25% Of A Webpage

For example, take this result from our data set.

Google Lens – Result From Data Set

If you look at the page this image is pulled from, you can see that the image appears quite high on the page (well above the fold).

Google Lens Example – Position On Web Page

This finding is in line with Google’s “Making visual content more useful in Search” blog post.

In that post they state that the algorithm “prioritizes” images high up on the page.

Google Algorithm Prioritizes Images High On Page

Our findings here are consistent with that statement.

Key Takeaway: Images that are higher up on a web page appear to have a significant ranking advantage in Google Lens.

A Fairly Low Number of Google Lens Result Images Contain Matching Alt Text

Google’s Search Engine Optimization Starter Guide states that they use alt text to understand images in their index.

Google Alt Text Makes It Easier

However, it’s not clear how important alt text is. Or whether it’s used in the Google Lens algorithm at all. Which is why we decided to see how often Google Lens result images had alt text that matched the visual search term that they ranked for.

We found that 11.4% of all Google Lens results contain alt text that “matches” the best-guess label for the search image.

Google Lens Results Contain Alt Text That Matches The Best Guess Label

For example, the best-guess label for this visual search is “yoga poses”.

Yoga Poses Example Search

And here are the results for that Google Lens search.

Google Lens – Yoga Image Results

In this case, 2 out of the 4 Google Lens results contain “yoga poses” in their alt text.

Yoga Pose Alt Text Overlay

11.4% isn’t an insignificant number. But it doesn’t suggest a strong relationship. If alt text was an important part of the Google Lens algorithm, we would expect to see more of the results to have exact match or partial match anchor text.

Key Takeaway: 11.4% of Google Lens image alt text match the search term that they rank for. Therefore, alt text may play a minimal role in Google Lens optimization.

Google Lens Results Have a High Page Authority and Domain Authority

It’s no secret that, all things being equal, an authority page on an authority site will outrank pages with less authority.

We wanted to know whether this same rule applied to Google Lens.

After all, Google themselves state that a recent visual search algorithm update made it so that authority pages had an edge in the results.

Google Blog – The Authority Of A Web Page

So we looked up Moz’s Domain Authority and Page Authority for each of our 65,388 Google Lens results.

And we found that Google Lens results tend to come from fairly authoritative sites and pages.

In fact, an average Google Lens result has a Page Authority of 35 and a Domain Authority of 64. Both fairly high.

An Average Google Lens Result Has A Page Authority Of 35 And A Domain Authority Of 64

It may be that Google directly uses link authority in the Google Lens algorithm. Or this relationship could simply be a case of “correlation not equaling causation”.

That’s because a high DA and PA tend to be byproducts of publishing high-quality content. And high-quality content is something that Google states is extremely important for visual search rankings.

In fact, in Google’s “Google Image best practices” documentation, they go as far as to say the quality of the content on the page is as important as the image itself.

Google Blog – Good Content Is Just As Important

Key Takeaway: We found that pages with a high Moz Domain Authority and Page Authority tend to rank well in Google Lens. This may be because Google Lens ranks pages with lots of backlinks. Or it could be that DA and PA are byproducts of publishing high-quality content.

The Vast Majority of Google Lens Results Come From Mobile-Friendly Sites

We found that 90.6% of all Google Lens image results are pages that pass Google’s Mobile-Friendly Test.

Google Lens Image Results Are Pages That Pass Googles Mobile Friendly Test

It makes sense that Google Lens would prefer to rank sites that are optimized for mobile. After all, essentially 100% of Google Lens searches are done on mobile devices.

And it would be a pretty poor user experience to send users to sites that didn’t work well on mobile, even if the image itself was a great result for the search.

In fact, Google’s recommendations for image SEO emphasize the importance of “device-friendly” web design.

Google Blog – Device Friendly Web Design

Key Takeaway: Approximately 77% of all pages with images ranking in Google Lens are mobile-friendly.

The Average First Contentful Paint Loading Speed of a Google Lens Result is 3,186ms

We decided to see if page loading speed played a role in Google Lens rankings.

For this analysis we measured First Contentful Paint (FCP) speed for our 65,388 Google Lens results.

We chose FCP for the simple reason that FCP is the time that many of a page’s assets (including images) appear. This is likely more important to someone conducting a visual search than Time to First Byte (TTFB), which doesn’t affect what users see on a page.

And we found that the average FCP for a Google Lens result was 3,186ms.

The Average FCP For A Google Lens Results In 3186ms

Based on recent page speed benchmarks, a 3,000+ FPC can be considered somewhat slow. However, this finding needs to be taken into context.

Pages that rank well in visual search tend to be image-heavy. Which adds significant size to any page.

In fact, we found that the average Google lens result contains an average of 34 total images.

The Average Google Lens Result Contains An Average 34 Total Images

That’s a lot of images on a single page. And when you take that fact into account, an image-heavy page that takes 3,000ms to reach FCP isn’t so bad.

While Google hasn’t said whether speed plays a role in their visual search algorithm, they do recommend that people that want to rank in Google Images should speed things up.

Google Blog – Optimize For Speed

So it could be that Google uses page speed as a visual search ranking signal. And that, due to the image-heavy nature of the pages that appear in visual search, 3,186ms puts a page into the top tier in terms of FCP.

Key Takeaway: Google Lens results don’t load significantly faster than the average webpage. In fact, the average Google Lens result FPC load time is 3,186ms. This may be due to the fact that visual search results tend to contain several high-quality images.

Pinterest and Amazon Rank For a Large Number of Visual Searches

Next, we decided to look at the distribution of websites that appear in Google Lens results.

Here’s the full breakdown:

Pinterest And Amazon Rank For A Large Number Of Visual Searches

As you can see, Pinterest and Amazon rank significantly more often than any other site.

In fact, 1/10 Google Lens results come from these two sites alone.

This finding aligns with a lot of our findings thus far. and both have a high Moz Domain Authority (96 and 94 respectively).

They also have fast loading times. And are built to be mobile friendly.

Importantly, both sites tend to have images appear at the very top of the page (usually above the fold).

Amazon Pinterest – Home Pages

However, outside of these outliers, Google Lens results are fairly diverse. 80.2% of Google Lens results are from the “other” category. Which means that no single set (or set of sites) completely dominates Google Lens results.

Key Takeaway: Pinterest and Amazon tend to rank most often Google Lens results. 11.3% of all visual search results are from these two websites.

Relatively Few Google Lens Images Are Responsive

Unlike some of the other factors that we analyzed, Google hasn’t stated that responsive images play a role in their visual search algorithm. But they do consider responsive images a Google Image SEO best practice because they “lead to better user experience”.

Google Blog – Responsive Images

While that may be the case, we didn’t find a strong relationship between responsive images and visual search rankings.

In fact, only 13% of the results in Google Lens are responsive.

Only 13% Of The Results In Google Lens Are Responsive

Key Takeaway: While responsive images may improve UX, they don’t appear to directly influence visual search.

Descriptive Image Filenames May Play a Role in Google Lens Results

Filenames have been an important Google Images ranking factor since its inception. Before Google gained the ability to recognize images with AI, they blindly trusted alt text and image filenames to understand the content of an image.

However, despite Google’s image recognition technology coming a long way, they still rely on filenames.

Google Blog – Image File Names

Which may explain the fact that a high number of images in the visual search results (22.6%) have a filename that aligns with what the person searched for in Google Lens.

Percentage Of Results That Have A Filename That Aligns With What The Person Searched For In Google Lens

For example, this Google Lens result has a filename (astronaut-deep-space-image-science-fiction-fantasy-in-high-resolution-picture) that closely matches one of the Google Vision labels (“space astronaut”) for the search image.

Astronaut example search

Key Takeaway: Google may use keywords that appear in an image’s filename as part of the Google Lens algorithm.

Google Lens Results Often Also Rank in Google Organic

Google’s visual search algorithm takes the relevancy of the page and website into account.

Here’s an example that Google recently gave to explain how site relevancy impacts visual search results:

Google Blog – DIY Search

We wanted to understand how this potentially important ranking factor may impact Google Lens searches.

Unfortunately, there’s no way to “look up” the relevancy of a website. But we can use a site’s organic Google rankings as a proxy measurement of relevancy.

For example, a site ranking highly in Google’s organic search results for the keyword “organic gardening” is considered by Google to be a relevant result for “organic gardening”.

And that may give that site a visual search SEO edge for searches related to organic gardening.

Indeed, that’s what the data in our analysis showed.

15% of the websites that rank in Google Lens also rank in Google’s organic top 10 results for that same term.

15% Of Websites That Rank In Google Lens Also Rank In Googles Organic Top 10 Results

For example, look at the #1 result in Google Lens when searching with an image of a mouse:

Google Lens – Mouse Search

That result comes from

Google Vision describes this image as “common california mice”.

Sure enough, when you search for “common california mice” in Google organic, ranks #1 for that same term:

Common California Mice

Key Takeaway: 15% of all Google Lens results also rank on the first page of Google’s “normal” organic search results.

The Average Word Count of a Google Lens Result Page is 1,631 Words

Our analysis found that Google Lens result pages tend to be fairly text-heavy. In fact, Google Lens images come from pages that contain an average of 1,631 words of text.

The Average Word Count Of A Google Lens Result Page Is 1631 Words

Why the relationship?

Well, Google officially recommends that “images are placed near relevant text”.

Google Blog – Images Placed Near Relevant Text

That’s because Google uses the text around an image to understand the content and context of that image.

Google Uses The Text Around An Image To Understand The Context Of That Image

Therefore, a page with considerable supporting text is going to have an advantage in visual search compared to a page that’s made up of 100% images.

Key Takeaway: Pages with a large amount of supporting text may perform best in visual search. Google Lens result pages contain an average of 1,631 words.


How Alaska Airlines leveraged search data to fill empty seats

For Alaska Airlines, growing routes by 40% in one year and becoming the airline with the most nonstop flights to and from the West Coast meant filling hundreds of new routes that had lower brand recognition. Owen Bickford, paid search program manager, shares the search data strategy that helped the airliner get it done.

After acquiring an airline and adding even more new routes in 2016, we knew we needed a fresh approach to marketing Alaska Airlines to new audiences. Our new combined network, numbering thousands of potential routes, offered a prime opportunity to revisit our marketing strategies with data.

Knowing search volume can be a proxy for customer demand, we partnered with Google to harness flight-specific searches. We wanted to uncover new areas of opportunity for selling empty seats via our paid search campaign efforts. Here’s how we did it.

We unified our search data to better identify growth opportunities

We worked with Google Flights to develop a common dataset to mine for insights. The result was a route coverage dashboard, an amalgam of Google Flights queries and Google Ads data, which we could filter in a multitude of ways. For example, we use the dashboard to identify specific routes where queries are high, but our ad impression and click coverage is low. That helps us decide if we want to invest more in those areas. The dashboard also quickly became a go-to resource for assessing competitive share and coverage gaps. As a result, we’re now able to make strategic shifts in our marketing campaigns to address previously uncovered routes.

The new strategy has garnered nearly 60 hours per year in time savings and fostered collaboration among our teams.

We automated our search campaigns

Armed with these new insights, our paid search team needed a way to manage search campaigns in an optimized way. We started using Search Ad 360’s upgraded inventory management, a tool that lets advertisers feed dynamic data into the platform for the automatic creation of keywords and ads. For us, that means generating highly targeted campaigns aligned with Alaska Airlines’ specific routes, allowing us to easily reallocate search budgets based on capacity and demand. We can now reliably manage paid search spend and respond quickly to changing market dynamics (such as new routes and flight availability) within hours, not days or weeks. The new workflow allowed us to build search coverage across thousands of routes, which drove a 69% increase in revenue and an 8% increase in return on ad spend.

We recognized full business impact to drive collaboration efforts

In addition to driving business results, the new strategy has garnered nearly 60 hours per year in time savings and fostered collaboration among our teams. We’re able to plan advertising at a route level, redistributing and increasing budgets as needed to increase the number of seats filled across our network.

Perhaps what’s most exciting is what we can do next: We’re now exploring how we can feed the tools additional data to better inform the output. And the collaboration enabled by the new workflow means teams such as marketing and revenue management are now synchronized to support routes, a major win for us in breaking down internal silos and better serving our customers.


Inside Google Marketing: Do’s and don’ts for marketing measurement during a pandemic

How do you measure your marketing efforts in a time of upheaval? It’s a pretty stark question, and one I’ve been asking myself a lot lately. As the head of strategic analytics for Google Marketing, I’ve been wrestling with how and what to measure since the COVID-19 pandemic upended, well, everything.

The pandemic has changed the personal and professional lives of everyone around the globe. Those reading this likely work in marketing. You may be reading from home or another remote location, juggling family commitments or home schooling, while dealing with the fact that entire campaigns have been suspended or drastically altered.

Should you even be measuring your marketing efforts during a time like this?

If the job of measuring marketing campaigns typically falls to you or your team, where do you even begin? And should you even be measuring your marketing efforts during a time like this?

It’s a challenge that we face at Google as well. To that end, we’ve identified five marketing measurement strategies we’re pausing for the moment, and five that continue to provide value.

We’ve put the brakes on:

  • Matched market tests. Comparing the behavior of users in a single control region with the behavior of users in a single test region requires stability. As regions, states, and geographies go in and out of social distancing protocols, with multiple changes in user behavior, stability will be heavily impacted. We’re also putting a halt to geo-experiments and ROPO (research online, purchase offline) tests for the same reason.
  • Short-term campaign KPIs. You’re going to have to make some tough calls about your long-term business objectives. If those objectives are still relevant, you might be tempted to change some of your long-term key performance indicators (KPIs) to focus only on short-term KPIs. We’re resisting that urge. Because these circumstances are so unique, you might not hit any of those short-term KPIs. It’s more than likely, anything you learn from the success or failure of short-term KPIs won’t be usable in the future.
  • Major strategic projects. We’ve seen some major changes in consumer media habits as well as responses to those shifts. There’s been an increase in the consumption of online news and linear TV. Meanwhile, increased demands on streaming services have led some providers to temporarily reduce video quality in an effort to reduce bandwidth usage. Whether these habits are long-lasting or temporary remains to be seen, but this is not the time to build learnings related to media approaches in a post-COVID world.
  • Face-to-face measurement. For the safety of all involved, we’ve requested our agencies and vendor partners to stop any and all face-to-face interviews, including exit interviews, market research, and in-person creative testing.
  • Unrealistic timelines. If a business-critical timeline is driving a campaign launch, proceed without worrying about the optimal measurement of that campaign. Optimal measurement always requires extra time to plan, evaluate, and implement necessary instrumentation. Let your executives know that now might not be the best time for such an approach. But if you absolutely have to strive for optimal measurement of a major campaign, be sure to give yourself more time in light of the current situation.

This could be the perfect time to invest in planning and upgrading your analytics strategies for 2021 and beyond.

We are continuing to:

  • Measure critical campaigns and channels. Clearly, you can’t — and shouldn’t stop — all measurement. Even in tough times, ensuring that you’re investing responsibly remains important. Which is what we’re doing. For major campaigns, we will continue to deploy the corresponding analytical elements across our strategic measurement stack. For campaigns with material budgets, we are applying preflight data checks on our media plans, analyzing live results to do in-flight optimization, and doing thorough post-campaign analysis.
  • Leverage remote creative pretesting. It is important that you understand how your ads are being perceived tonally, especially during times like these. One misstep could have a lasting impact on your brand image. For new creative, normal pretesting remains crucial for us. It’s proven time and again to be a strong predictor of in-market performance. We even recommend retesting creative that was in use before COVID-19 to ensure it is still effective and relevant — and that it doesn’t run the risk of striking the wrong note.
  • Focus on strategic, cross-marketing meta-analysis. Your business and ours moves really fast. As a result, analysts get caught in the “trees view” of data. Taking advantage of this opportunity, our team is scaling up our efforts to look at the “forest views” hidden in our data sets. The sky’s the limit here, but a couple of things I’m focusing on are effectiveness of digital channels in delivering value across marketing initiatives and clear cross-channel cause and effect of various tactics we deploy. I also obsess about in-flight signals, so we are looking for new ones to feed our algorithmic optimization efforts.
  • Take advantage of think time. Pausing a lot of the things you usually do means that you may have something you usually find in short supply: time to think. This could be the perfect time to invest in planning and upgrading your analytics strategies for 2021 and beyond. None of us knows what life on the other side of COVID-19 will look like. But for now we can try to figure out what our best practices, guidelines, and guardrails for measurement should look like in the future.
  • Invest in structural upgrades. Perhaps the past few years have found you working at such a pace that you’ve continually put off much-needed upgrades. If you’re seeing a reduction in your large-scale campaign measurement efforts, now might be the time to put in place structural upgrades to your analytical capabilities. On top of doing that, we’re working with our agency partners to optimize and streamline reporting and analysis cadences across touchpoints.

This isn’t an easy time for anyone. Like everyone else, we hope this is a rare occurrence and that the period of disruption will end sooner rather than later. But we still have to deal with the reality of the present. As marketing reacts to this ever-shifting landscape, measurement strategies should be similarly agile.